ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-28 02:11:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

bitnet: add 2 bpw quantization

Browse Source 
The scalar dot product already chieves 37 t/s for TG!
This commit is contained in:
Iwan Kawrakow
2024-06-17 18:41:30 +03:00
parent 765622ff8f
commit f6863cfa1b
10 changed files with 216 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
examples/quantize-stats/quantize-stats.cpp
View File

@@ -341,6 +341,10 @@ int main(int argc, char ** argv) {
if (!layer_included(params, kv_tensor.first)) {
continue;
}
if (kv_tensor.second->ne[0] == 1 || kv_tensor.second->ne[1] == 1) {
// we never quantize those
continue;
}
if (params.verbose) {
printf("%s: type %s, size %" PRId64 "\n", kv_tensor.first.c_str(), ggml_type_name(kv_tensor.second->type), ggml_nelements(kv_tensor.second));
}
@@ -386,6 +390,10 @@ int main(int argc, char ** argv) {
if (!layer_included(params, kv_tensor.first)) {
continue;
}
if (kv_tensor.second->ne[0] == 1 || kv_tensor.second->ne[1] == 1) {
// we never quantize those
continue;
}
if (params.verbose) {
printf(" %s ...\n", kv_tensor.first.c_str());
}

1
examples/quantize/quantize.cpp
View File

@@ -27,6 +27,7 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S, " 1.56 bpw quantization", },
{ "IQ1_M", LLAMA_FTYPE_MOSTLY_IQ1_M, " 1.75 bpw quantization", },
{ "IQ1_BN", LLAMA_FTYPE_MOSTLY_IQ1_BN, " 1.75 bpw quantization (Bitnet)", },
{ "IQ2_BN", LLAMA_FTYPE_MOSTLY_IQ2_BN, " 2.00 bpw quantization (Bitnet)", },
{ "Q2_K", LLAMA_FTYPE_MOSTLY_Q2_K, " 2.63G, +0.6717 ppl @ LLaMA-v1-7B", },
{ "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S, " 2.16G, +9.0634 ppl @ LLaMA-v1-7B", },
{ "IQ3_XXS",LLAMA_FTYPE_MOSTLY_IQ3_XXS," 3.06 bpw quantization", },